[mvapich-discuss] Troubles starting simple 2 process IMB-MPI1 using mvapich2-2.1

[Devesh Sharma]

Hi Hari,

Thanks for your quick help, changing the env to MV2_USE_RoCE helped and it
starting working.

On Mon, Apr 20, 2015 at 8:52 PM, Hari Subramoni <subramoni.1 at osu.edu> wrote:

> Hello Devesh,
>
> MVAPICH2 does not support Emulex HCAs at this point in time.
>
> One side note - the environment variable for IB RoCE is MV2_USE_RoCE not
> MV2_USE_RDMAOE.
>
> Best Regards,
> Hari.
>
> On Mon, Apr 20, 2015 at 11:01 AM, Devesh Sharma <devesh28 at gmail.com>
> wrote:
>
>> Hello list,
>>
>> I am trying to run IMB 4.0.2 with latest MVAPICH2-2.1 release with Emulex
>> RoCE device. I have set the ulimts to the unlimited. However I am seeing
>> that my job is stuck, neither its aborting and coming out to the shell nor
>> it is running.
>>
>> I tried attaching gdb on the other node, and found that its stuck
>> somewhere in the exit path, following is the stack trace
>>
>> (gdb) bt
>> #0  0x00002adae3df04f6 in malloc_consolidate () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #1  0x00002adae3df2003 in _int_malloc () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #2  0x00002adae3df2d0a in malloc () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #3  0x00002adae41b68bd in __fopen_internal () from /lib64/libc.so.6
>> #4  0x00002adae41f99dd in __tzfile_read () from /lib64/libc.so.6
>> #5  0x00002adae41f8d24 in tzset_internal () from /lib64/libc.so.6
>> #6  0x00002adae41f96d3 in __tz_convert () from /lib64/libc.so.6
>> #7  0x00002adae3d96207 in MPID_Abort () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #8  0x00002adae3d6366f in handleFatalError () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #9  0x00002adae3d63771 in MPIR_Err_return_comm () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #10 0x00002adae3d1efc4 in PMPI_Init () from
>> /usr/local/mpi/mvapich2/lib/libmpi.so
>> #11 0x00000000004018f9 in main ()
>> (gdb) q
>> A debugging session is active.
>>
>>         Inferior 1 [process 7099] will be detached.
>>
>> Quit anyway? (y or n) y
>>
>>
>> Following is LDD output:
>>
>> -bash-4.2# ldd /usr/local/imb/mvapich2/IMB-MPI1
>>         linux-vdso.so.1 =>  (0x00007fff5d5fe000)
>>         libmpi.so.12 => /usr/local/mpi/mvapich2/lib/libmpi.so.12
>> (0x00002b1bdf26d000)
>>         libc.so.6 => /lib64/libc.so.6 (0x00002b1bdf904000)
>>         libibmad.so.5 => /lib64/libibmad.so.5 (0x00002b1bdfcc5000)
>>         librdmacm.so.1 => /lib64/librdmacm.so.1 (0x00002b1bdfedf000)
>>         libibumad.so.3 => /lib64/libibumad.so.3 (0x00002b1be00f6000)
>>         libibverbs.so.1 => /lib64/libibverbs.so.1 (0x00002b1be02ff000)
>>         libdl.so.2 => /lib64/libdl.so.2 (0x00002b1be050d000)
>>         librt.so.1 => /lib64/librt.so.1 (0x00002b1be0711000)
>>         libgfortran.so.3 => /lib64/libgfortran.so.3 (0x00002b1be0919000)
>>         libm.so.6 => /lib64/libm.so.6 (0x00002b1be0c3a000)
>>         libpthread.so.0 => /lib64/libpthread.so.0 (0x00002b1be0f3c000)
>>         libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00002b1be1158000)
>>         libquadmath.so.0 => /lib64/libquadmath.so.0 (0x00002b1be136e000)
>>         /lib64/ld-linux-x86-64.so.2 (0x00002b1bdf04a000)
>> -bash-4.2#
>>
>> Command Used:
>>
>> /usr/local/mpi/mvapich2/bin/mpirun --verbose -np 2 -f /root/hostfile -env
>> MV2_USE_RDMAOE 1 -env MV2_USE_RDMA_CM 1 -env MV2_DEFAULT_MAX_CQ_SIZE 4096
>> -env MV2_USE_SRQ 0 -env MV2_MX_SEND_WR 1024 -env MV2_MAX_RECV_WR 1024 -env
>> MV2_USE_SHARED_MEM 1 -env MV2_USE_BLOCKING 0  -env MV2_USE_UD_HYBRID 0 -env
>> MV2_VBUF_TOTAL_SIZE 65536 -env MV2_IBA_EAGER_THRESHOLD 65536 -env
>> MV2_IBA_HCA ocrdma2 /usr/local/imb/mvapich2/IMB-MPI1
>>
>> I have also tired workaround specified in Section 9.1.3 of user guide but
>> no help. The same command with same mvapich2 version runs fine on another 2
>> node cluster. Any pointers to resolve this issue would be a great help.
>>
>>
>> --
>> Please don't print this E- mail unless you really need to - this will
>> preserve trees on planet earth.
>>
>> _______________________________________________
>> mvapich-discuss mailing list
>> mvapich-discuss at cse.ohio-state.edu
>> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>>
>>
>


-- 
Please don't print this E- mail unless you really need to - this will
preserve trees on planet earth.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20150420/a2af1a8d/attachment-0001.html>